CN101576956B - On-line character detection method based on machine vision and system thereof - Google Patents
On-line character detection method based on machine vision and system thereof Download PDFInfo
- Publication number
- CN101576956B CN101576956B CN2009103022001A CN200910302200A CN101576956B CN 101576956 B CN101576956 B CN 101576956B CN 2009103022001 A CN2009103022001 A CN 2009103022001A CN 200910302200 A CN200910302200 A CN 200910302200A CN 101576956 B CN101576956 B CN 101576956B
- Authority
- CN
- China
- Prior art keywords
- character
- detection
- similarity
- type matrix
- checked
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
- Character Input (AREA)
Abstract
The invention belongs to the technical field of automatic detection and relates to an on-line character detection method based on machine vision. The detection method comprises a plurality of softwaremodules including an image pre-treatment software module, a target extraction software module, a type matrix extraction software module, a character sequence mode generation software module, a charac ter detection software module and so on. The above modules are combined together to form two independent flows including a production establishment operation flow and a character real-time detection flow. The invention synchronously provides an on-line detection system for realizing the detection method. The detection system and method provided in the invention have strong robustness, are capable of effectively reducing the influence of image variations, such as brightness, contrast, quality, character stroke, rotation, scale and the like, caused by environment or other factors, and are capable of realizing fast and correct character region location and high-efficiency and real-time charter on-line detection.
Description
Technical field
The invention belongs to computer vision, Flame Image Process and mode identification technology, relate to a kind of online character detection method and system.
Background technology
Along with manufacturing enterprise's scale, informationalized continuous development, manufacturer is more and more higher to the requirement that characters such as online Product labelling detect, and need realize the detection of pin-point accuracy on production line at a high speed.Rely on the method for manual detection can't be competent at, the application of the Machine Vision Detection of high-speed, high precision can significantly be improved the quality of products and reliability at all.It is the character that utilizes on automatic identification Product labelling of computing machine or the packing that character detects, and is an important branch of pattern-recognition, belongs to the category of character recognition technologies.Character recognition technologies development in these several years has broad application prospects at aspects such as logistics information, meter reading, car plate identification, bill identification, I.D. discriminatings, to reach the purpose of raising the efficiency and reducing cost of labor rapidly.At present, the recognition methods of character mainly contains the method based on template matches, based on the method for charcter topology with based on neural network method, these existing image-recognizing methods all more or less have certain limitation, under a kind of environment the good method of effect change a kind of environment recognition effect may be very undesirable.Some have certain versatility, the method that recognition effect is good, and often calculated amount is very big, is difficult to real-time application.
Summary of the invention
The problem that can't handle in real time for the deficiency that overcomes single recognition methods detectability in the prior art and universal method, the invention provides a kind of online character detection method and detection system, detection method and the detection system of utilizing the present invention to propose can realize quick, high precision character measuring ability.
It is as follows that the present invention solves the scheme that its technical matters adopts:
A kind of online character detection method based on machine vision comprises the following steps, (1) wherein to (9) step is that product is set up operating process, and step (10) to (14) is the online testing process of character:
(1) reference picture that comprises character that collects is carried out pre-service;
(2) in through the reference picture that obtains after the pre-service, select to have the part of rigidity characteristic as localizing objects in character zone or the close position;
(3) in localizing objects, extract the closed contour feature and the line feature of non-closure, the geometric properties that extracts is sorted according to scale size;
(4) according to from big to small order each yardstick is made as segmentation threshold, the geometric properties that is extracted is divided into large-scale characteristics collection and two parts of small scale features collection, in reference picture, carry out overall situation characteristic matching fast by the large-scale characteristics collection, on the basis of coarse positioning, utilize the small scale features collection to carry out the local feature coupling again with accurate localizing objects, the matching precision and the speed of the each location of record, determine the weighting coefficient of each scale feature according to matching speed, be chosen in the yardstick of obtaining maximum balance between matching speed and the matching precision and cut apart, localizing objects is expressed as the data structure of two characteristic sets of size and characteristic weighing coefficient formation as the best scale threshold value and to the geometric properties that extracts;
(5) from reference picture, extract type matrix, and specify the true origin of type matrix;
(6) by being set, length, width, name and type matrix initial point define the empty word mould;
(7) pixel value of type matrix is made amendment, improve type matrix subgraph details, and carry out mask process;
(8) seven Hu invariant moment features of extraction type matrix adopt formula
Calculate the similarity between the type matrix and generate the coefficient of similarity matrix, in the formula, (u v) is searching position (u, the invariant moments correlation of v) going up, M to R
iBe match map invariant moments, N
i(u, v) reference diagram searching position (u, the invariant moments that v) goes up;
(9) generate the character string pattern, wherein comprise: the character to be detected of corresponding matrix magazine, composition sequence, the true origin of character string pattern; Each character is based on the detection acceptance threshold of each character in the relative coordinate of pattern initial point, the sequence with put the letter threshold value in the sequence;
(10) set the operational factor of online character detection system, and utilize this device to gather the image that comprises character to be checked in real time;
(11) image that comprises character to be checked that collects is carried out the image pre-service;
(12) in the region of search of setting, in specifying the region of search, carry out coarse positioning by large-scale characteristics collection and related weighing coefficient, by small scale features collection and related weighing coefficient target is accurately located on this basis, adjust type matrix image and the sub-piece of matching image according to the operational factor that sets and repeat above-mentioned coarse positioning and accurate position fixing process, select with extract target have maximum similarity and this value above the image-region of locating acceptance threshold as positioning result;
(13), global coordinate system is converted to the initial point of the localizing objects local coordinate system as initial point according to positioning result;
(14) carry out the detection of static character string, dynamic character sequence and null character (NUL).
As preferred implementation, above-mentioned online character detection method in the step (3), utilizes Canny operator extraction feature; The operational factor of setting in the step (10) comprises target localization parameter and character detected parameters, wherein the target localization parameter comprises and sets location acceptance threshold, region of search and target rotates, the permissible range of dimensional variation, the character detected parameters comprise set that each character rotates in the character string to be checked, the permissible range of translation, dimensional variation.
The detection of the static character string in the step (14) can be undertaken by following steps:
The first step, according to each character in positioning result and the sequence based on the relative coordinate of pattern initial point, the type matrix of choosing image subblock and character to be checked in the target location carries out invariant moments normalization relevant matches and calculates, obtain similarity, adjust matched position according to default operational factor and mate continuously, if maximum similarity is less than detecting acceptance threshold, then this character detects failure, otherwise, continue to carry out next step;
In second step,, determine in this matrix magazine similar to character height to be checked and may cause the easy gibberish symbol of erroneous judgement according to type matrix compiling gained coefficient of similarity matrix and default similarity threshold;
In the 3rd step, calculate the similarity that all easy gibberish accord with according to the described process of the first step;
In the 4th step, if the difference between the maximum similarity of the maximum similarity of character to be checked and easy gibberish symbol is put the letter threshold value greater than detection, then this character detects successfully; Otherwise detect failure.
The detection of the dynamic character sequence in the step (15) is undertaken by following steps:
The first step, according to each character in positioning result and the sequence based on the relative coordinate of pattern initial point, the type matrix of choosing image subblock and character example to be checked in the target location carries out invariant moments normalization relevant matches one by one and calculates, obtain similarity, and the default operational factor adjustment matched position of foundation mates continuously, if the maximum similarity of arbitrary character or character example all is lower than the detection acceptance threshold, then detect failure;
In second step, get character with maximum similarity or character example as character to be checked;
In the 3rd step,, in this matrix magazine, determine similar and may cause that the easy gibberish of erroneous judgement accords with to character height to be checked according to type matrix compiling gained coefficient of similarity matrix and set similarity threshold.
In the 4th step, calculate the similarity that all easy gibberish accord with according to the described process of the first step.
In the 5th step, if the difference between the maximum similarity of the maximum similarity of character to be checked and easy gibberish symbol is put the letter threshold value greater than detection, then this character detects successfully; Otherwise detect failure.
Step (15) is carried out the detection of null character (NUL) by following steps after every other nonblank character detection finishes: determine by the average gray value of null character (NUL) more to be checked and the average gray value of its adjacent character whether null character (NUL) exists earlier, calculate the null character (NUL) matching degree again and compare, judge with this whether null character (NUL) detects success with detecting acceptance threshold.
The present invention provides a kind of system that realizes above-mentioned online character detection method based on machine vision simultaneously, comprise video camera, light source, image pick-up card and computing machine, described video camera is used for the image that online acquisition comprises character, the image that is collected is sent to computing machine through image pick-up card, described light source adopts the annular light source that surrounds video camera, contain at described calculator memory reference picture is carried out pre-service, target is extracted and type matrix extracts data and the character string mode data that the back generates, and store image is carried out pre-service, target localization and character detect software, are used for static character string, dynamic character sequence and null character (NUL) are carried out online detection.
Detection system provided by the invention and method, integrated application the technology of Flame Image Process, pattern-recognition, field of machine vision, effectively extract the characteristic information of target and character, on the basis of target localization, detect, has very strong robustness, the influence of the image change such as brightness, contrast, quality, character stroke, rotation and yardstick that environment or other factors cause can effectively reduce owing to can realize character zone location, the efficient real-time online detection of character fast and accurately.
Description of drawings
Fig. 1: hardware system block diagram of the present invention;
Fig. 2: software product of the present invention is set up operational flowchart;
Fig. 3: the real-time testing process figure of software character of the present invention;
Among Fig. 1: 1 video camera, 2 light sources, 3 transmission lines, 4 image pick-up cards, 5 computing machines.
Embodiment
The present invention is a kind of universal character detection system, and this hardware system is made up of hardware components and software section.
As shown in Figure 1, this hardware components is made up of video camera, light source, transmission line, image pick-up card and computing machine.The camera lens of present embodiment adopts the line array CCD camera.Light source adopts the annular light source that surrounds video camera.Image pick-up card adopts the outer plug-in card that has the image front-end processing based on microcomputer pci bus structure, with more complete real time image collection disposal system of existing resource formation of microcomputer.Camera acquisition realtime graphic, size are 800 * 600, by transmission line and image pick-up card deliver to Computer Storage be the BMP form and and detect in real time.
Character detection system of the present invention is made up of hardware components and software section; Hardware components adopts computing machine as handling and control center; Software section is made up of a plurality of modules such as image pre-service, target extraction, type matrix extraction, character string mode producing, character detections.In actual applications, above-mentioned module combinations gets up to form two independently flow processs: product is as shown in Figure 2 set up operating process and the real-time testing process of character as shown in Figure 3.Below software systems are described in detail.
1) the image pretreatment module comprises: image gray processing, expeling noise, grey level stretching, binaryzation and morphologic filtering.
Image gray processing, the image that camera is taken in real time is 16 bitmaps (RGB565 form) data, handles for the ease of follow-up rapid image, needs view data is changed, and makes coloured image become 256 grades of gray-scale maps.
Remove noise, inevitably contain noise in the image, adopt medium filtering that image is carried out pre-service thereupon.
Grey level stretching in order to strengthen the contrast of background area and character zone, is carried out grey level stretching to image.
Binaryzation is carried out binary conversion treatment to gray level image, adopts the method for maximum between-cluster variance and infima species internal variance ratio, and self-adaptation is calculated gray threshold, thinks the target area less than the zone of this threshold value, greater than the background area of thinking in the zone of this threshold value.
Morphologic filtering carries out morphologic filtering to bianry image and handles, the synthetic operation that adopts expansion, burn into ON operation and closed operation to combine.
2) target extraction module includes: target selection, feature extraction, location training.
Target selection selects character zone or close position to have the part of remarkable rigidity characteristic as localizing objects in reference picture.
Feature extraction utilizes the Canny operator to extract the closed contour feature and the line feature of non-closure in localizing objects, according to scale size the geometric properties that extracts is sorted.
The location training, according to from big to small order each yardstick is made as segmentation threshold, the geometric properties that is extracted is divided into large-scale characteristics collection and two parts of small scale features collection, in reference picture, carry out overall situation characteristic matching fast by the large-scale characteristics collection, on the basis of coarse positioning, utilize the small scale features collection to carry out the local feature coupling again with accurate localizing objects, the matching precision and the speed of the each location of record, determine the weighting coefficient of each scale feature according to matching speed, being chosen in the yardstick of obtaining maximum balance between matching speed and the matching precision cuts apart as the best scale threshold value and to the geometric properties that extracts, object table is shown as the data structure of two characteristic sets of size and characteristic weighing coefficient formation, as the basis of succeeding target location, the speed of target localization when training process can be accelerated the character detection.
3) the type matrix extraction module includes: common type matrix extraction, the definition of empty word mould, type matrix editor, type matrix compiling.
Type matrix extracts, extract character picture and name from reference picture, a plurality of type matrixes with same names are called a plurality of examples of this character, and specify the true origin of type matrix, generally be made as the geometric center of type matrix, reduce the coupling calculated amount when rotating with convergent-divergent with convenient character.
The definition of empty word mould defines the empty word mould by length, width, name and type matrix initial point are set, and to strengthen the dirigibility that character detects, different with common type matrix, the gap mould does not comprise any image pixel.
The type matrix editor makes amendment to the pixel value of type matrix image, improves type matrix subgraph details; The type matrix image that extracts is carried out mask process, and for example virtualization or the emergence edge to type matrix carries out mask, and then follow-up type matrix compilation process is not extracted the character feature of mask part, helps improving speed and the precision that character detects.
The type matrix compiling, extract seven HuShi invariant moment features of type matrix, adopt invariant moments normalization related algorithm as shown in Figure 1 to calculate the similarity between the type matrix and generate the coefficient of similarity matrix, the coefficient of similarity matrix is used for getting rid of when character detects highly similar and may causes that the character of erroneous judgement disturbs.
Wherein, (u v) is searching position (u, the invariant moments correlation of v) going up, M to R
iBe match map invariant moments, N
i(u, v) reference diagram searching position (u, the invariant moments that v) goes up.In above-mentioned type matrix similarity is calculated, at first two type matrixes are carried out normalized, thereby (u, v) value (0,0).
4) character string mode producing module comprises: the character string pattern is provided with, the character string pattern drill.
The setting of character string pattern comprises simple mode setting and fine mode setting.Simple mode is provided with specifies changeless character string to be detected; be corresponding fixing all examples of type matrix or same type matrix of each character position; set the detection order (left side trend or right trend) and the training zone of character string, and set the true origin of character string pattern and the relative position of sequence of characters.Fine mode is provided for detecting the character string of dynamic change, as the product batch number or the date of increasing or decreasing, the key distinction of itself and simple mode setting is, the character position of character string to be checked can comprise a plurality of characters or only comprise some examples of same type matrix, has improved the dirigibility that detects.
The character string pattern drill is provided with content and corresponding type matrix comprehensively becomes special data structure with aforementioned, is called the character string pattern, wherein comprises: corresponding matrix magazine; The character to be detected of composition sequence; The true origin of character string pattern; Each character is based on the relative coordinate of pattern initial point in the sequence; The detection acceptance threshold of each character and put the letter threshold value in the sequence.
5) character detection module comprises: operating parameter setting, target localization, coordinate conversion and character detect.
Operating parameter setting, target localization parameter and character detected parameters are set, wherein the target localization parameter comprises that acceptance threshold, region of search (being defaulted as entire image) are located in setting and target rotates, the permissible range of dimensional variation, the character detected parameters comprise set that each character rotates in the character string to be checked, the permissible range of translation, dimensional variation, wherein the permissible range of target and dimensional variation is limited between positive and negative 0.1.
Target localization, training result according to the target extraction module, in the region of search of setting, in specifying the region of search, carry out coarse positioning by large-scale characteristics collection and related weighing coefficient, by small scale features collection and related weighing coefficient target is accurately located on this basis, adjust type matrix image and the sub-piece of matching image according to the rotation of setting and yardstick permissible range and repeat above-mentioned coarse positioning and accurate position fixing process, select with extract target have maximum similarity and this value above the image-region of locating acceptance threshold as positioning result.
Coordinate conversion according to positioning result, is converted to global coordinate system with the initial point of the localizing objects local coordinate system as initial point.
Character detects, and comprising: the detection of the detection of static character string, dynamic character sequence, the detection of null character (NUL).
The detection of static character string is undertaken by following steps:
The first step, according to each character in positioning result and the sequence based on the relative coordinate of pattern initial point, the type matrix of choosing image subblock and character to be checked in the target location carries out invariant moments normalization relevant matches and calculates, the gained result is a similarity, its value is between 0 to 1, the permissible range adjustment matched position that changes according to default translation mates continuously, and wherein the maximal value of similarity is called maximum similarity.If maximum similarity is less than detecting acceptance threshold, then this character detects failure.
In second step,, determine in this matrix magazine similar to character height to be checked and may cause the easy gibberish symbol of erroneous judgement according to type matrix compiling gained coefficient of similarity matrix and default similarity threshold.
In the 3rd step, calculate the similarity that all easy gibberish accord with according to the described process of the first step.
In the 4th step, if the difference between the maximum similarity of the maximum similarity of character to be checked and easy gibberish symbol is put the letter threshold value greater than detection, then this character detects successfully; Otherwise detect failure.
The detection of dynamic character sequence is undertaken by following steps:
The first step, according to each character in positioning result and the sequence based on the relative coordinate of pattern initial point, the type matrix of choosing image subblock and character example to be checked in the target location carries out invariant moments normalization relevant matches one by one and calculates, the gained result is a similarity, its value is between 0 to 1, and change permissible range according to default translation and adjust matched position and mate continuously, wherein the maximal value of similarity is called maximum similarity.If the maximum similarity of arbitrary character or character example all is lower than the detection acceptance threshold, then detect failure.
In second step, get character with maximum similarity or character example as character to be checked.
In the 3rd step,, in this matrix magazine, determine similar and may cause that the easy gibberish of erroneous judgement accords with to character height to be checked according to type matrix compiling gained coefficient of similarity matrix and set similarity threshold.
In the 4th step, calculate the similarity that all easy gibberish accord with according to the described process of the first step.
In the 5th step, if the difference between the maximum similarity of the maximum similarity of character to be checked and easy gibberish symbol is put the letter threshold value greater than detection, then this character detects successfully; Otherwise detect failure.
The detection of null character (NUL) detects null character (NUL) after every other nonblank character detection finishes.Earlier determine by the average gray value of null character (NUL) more to be checked and the average gray value of its adjacent character whether null character (NUL) exists, calculate the null character (NUL) matching degree again and compare, judge with this whether null character (NUL) detects success with the detection acceptance threshold.
Illustrate character of the present invention below and detect software systems.
Product is set up operating process: camera is gathered a frame as the reference image, calling graph as pretreatment module to reference picture strengthen, filtering and binary conversion treatment; In reference picture, select character zone or close position to have the part of remarkable rigidity characteristic as localizing objects, such as part in the green square frame shown in Figure 3, extract the geometric properties of this target, the control reference image is trained, and object table is shown as the data structure of a characteristic feature and weighting coefficient formation thereof and is saved in the product information file; The character extraction module extracts character picture, specify the true origin of type matrix, set similarity threshold, generally be made as about 0.5, and compiling generates corresponding with type matrix data structure and the similarity matrix of being made up of characteristic feature and weighting coefficient thereof, the type matrix image with compile the result and be saved in the product information file; Character string mode producing module, specify long word symbol sequence to be detected, if same character position is set a plurality of characters, then can be used to detect the dynamic character sequence, and be used to judge the detection acceptance threshold that detection is whether successful and put the letter threshold value that for each character member of sequence sets common acceptance threshold is set to about 0.5 one by one, putting the letter threshold value is set to about 0.2, generate the character string pattern by training, preservation mode data in the product information file, and detect test.Described product is set up flow process and is used for before character detects in real time each new product to be checked being set up its relevant product information hereof.
The character testing process: at first selected product based on the observed result that may change character zone and character, is provided with the correlation parameter that target localization and character detect by the operating parameter setting module; Utilize online character detection system of the present invention to gather the image that comprises character to be checked in real time; Calling graph as pretreatment module to input picture strengthen, filtering and binary conversion treatment; Data structure according to the storage target signature that generates in the target extraction module, in specifying the region of search, mate calculating, coupling based target shape rather than gray level pixel value, the coupling of arbitrary position need travel through the permissible range of default translation, rotation and dimensional variation, selects to have maximum similarity and this value and surpasses the zone of location acceptance threshold as positioning result; At locating area, the data structure that character detection module invokes character string mode producing module generates detects each character in the sequence to be checked, and export corresponding information, if an above character detects failure in the sequence to be checked, then preserve error image and relevant information, simultaneously output alarm signal.
Claims (6)
1. the online character detection method based on machine vision comprises the following steps, (1) wherein to (9) step is that product is set up operating process, and step (10) to (14) is the online testing process of character:
(1) reference picture that comprises character that collects is carried out pre-service;
(2) in through the reference picture that obtains after the pre-service, select to have the part of rigidity characteristic as localizing objects in character zone or the close position;
(3) in localizing objects, extract the closed contour feature and the line feature of non-closure, the geometric properties that extracts is sorted according to scale size;
(4) according to from big to small order each yardstick is made as segmentation threshold, the geometric properties that is extracted is divided into large-scale characteristics collection and two parts of small scale features collection, in reference picture, carry out overall situation characteristic matching fast by the large-scale characteristics collection, on the basis of coarse positioning, utilize the small scale features collection to carry out the local feature coupling again with accurate localizing objects, the matching precision and the speed of the each location of record, determine the weighting coefficient of each scale feature according to matching speed, be chosen in the yardstick of obtaining maximum balance between matching speed and the matching precision and cut apart, localizing objects is expressed as the data structure of two characteristic sets of size and characteristic weighing coefficient formation as the best scale threshold value and to the geometric properties that extracts;
(5) from reference picture, extract type matrix, and specify the true origin of type matrix;
(6) define the empty word mould by the true origin that length, width, name and type matrix are set;
(7) pixel value of type matrix is made amendment, improve type matrix subgraph details, and carry out mask process;
(8) seven Hu invariant moment features of extraction type matrix adopt formula
Calculate the similarity between the type matrix and generate the coefficient of similarity matrix, in the formula, (u v) is searching position (u, the invariant moments correlation of v) going up, M to R
iBe match map invariant moments, N
i(u, v) reference diagram searching position (u, the invariant moments that v) goes up;
(9) generate the character string pattern, wherein comprise: the character to be detected of corresponding matrix magazine, composition sequence, the true origin of character string pattern; Be the detection acceptance threshold of each character in the relative coordinate, sequence of each character in the determined sequence of benchmark with the true origin of described pattern and put the letter threshold value;
(10) set the operational factor of online character detection system, and utilize this device to gather the image that comprises character to be checked in real time;
(11) image that comprises character to be checked that collects is carried out the image pre-service;
(12) in the region of search of setting, in specifying the region of search, carry out coarse positioning by large-scale characteristics collection and related weighing coefficient, by small scale features collection and related weighing coefficient target is accurately located on this basis, adjust type matrix image and the sub-piece of matching image according to the operational factor that sets and repeat above-mentioned coarse positioning and accurate position fixing process, select with extract target have maximum similarity and this value above the image-region of locating acceptance threshold as positioning result;
(13), global coordinate system is converted to the initial point of the localizing objects local coordinate system as initial point according to positioning result;
(14) carry out the detection of static character string, dynamic character sequence and null character (NUL).
2. online character detection method according to claim 1 is characterized in that, in the step (3), utilizes Canny operator extraction feature.
3. online character detection method according to claim 1, it is characterized in that, the operational factor of setting in the step (10) comprises target localization parameter and character detected parameters, wherein the target localization parameter comprises and sets location acceptance threshold, region of search and target rotates, the permissible range of dimensional variation, the character detected parameters comprise set that each character rotates in the character string to be checked, the permissible range of translation, dimensional variation.
4. online character detection method according to claim 1 is characterized in that, the detection of the static character string in the step (14) is undertaken by following steps:
The first step, relative coordinate according to each character in positioning result and the sequence, the type matrix of choosing image subblock and character to be checked in the target location carries out invariant moments normalization relevant matches and calculates, obtain similarity, adjust matched position according to default operational factor and mate continuously, if maximum similarity is less than detecting acceptance threshold, then this character detects failure, otherwise, continue to carry out next step;
In second step,, determine in this matrix magazine similar to character height to be checked and may cause the easy gibberish symbol of erroneous judgement according to type matrix compiling gained coefficient of similarity matrix and default similarity threshold;
In the 3rd step, calculate the similarity that all easy gibberish accord with according to the described process of the first step;
In the 4th step, if the difference between the maximum similarity of the maximum similarity of character to be checked and easy gibberish symbol is put the letter threshold value greater than detection, then this character detects successfully; Otherwise detect failure.
5. online character detection method according to claim 1 is characterized in that, the detection of the dynamic character sequence in the step (14) is undertaken by following steps:
The first step, according to each character in positioning result and the sequence based on the relative coordinate of pattern initial point, the type matrix of choosing image subblock and character example to be checked in the target location carries out invariant moments normalization relevant matches one by one and calculates, obtain similarity, and the default operational factor adjustment matched position of foundation mates continuously, if the maximum similarity of arbitrary character or character example all is lower than the detection acceptance threshold, then detect failure;
In second step, get character with maximum similarity or character example as character to be checked;
In the 3rd step,, in this matrix magazine, determine similar and may cause that the easy gibberish of erroneous judgement accords with to character height to be checked according to type matrix compiling gained coefficient of similarity matrix and set similarity threshold.
In the 4th step, calculate the similarity that all easy gibberish accord with according to the described process of the first step.
In the 5th step, if the difference between the maximum similarity of the maximum similarity of character to be checked and easy gibberish symbol is put the letter threshold value greater than detection, then this character detects successfully; Otherwise detect failure.
6. online character detection method according to claim 1, it is characterized in that, step (14) is carried out the detection of null character (NUL) by following steps after every other nonblank character detection finishes: determine by the average gray value of null character (NUL) more to be checked and the average gray value of its adjacent character whether null character (NUL) exists earlier, calculate the null character (NUL) matching degree again and compare, judge with this whether null character (NUL) detects success with detecting acceptance threshold.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009103022001A CN101576956B (en) | 2009-05-11 | 2009-05-11 | On-line character detection method based on machine vision and system thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009103022001A CN101576956B (en) | 2009-05-11 | 2009-05-11 | On-line character detection method based on machine vision and system thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101576956A CN101576956A (en) | 2009-11-11 |
CN101576956B true CN101576956B (en) | 2011-08-31 |
Family
ID=41271889
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009103022001A Expired - Fee Related CN101576956B (en) | 2009-05-11 | 2009-05-11 | On-line character detection method based on machine vision and system thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101576956B (en) |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103093185B (en) * | 2011-11-07 | 2016-05-25 | 佳能株式会社 | Character recognition device, image processing apparatus and method thereof |
CN103646390B (en) * | 2013-09-27 | 2016-11-23 | 北京中钞钞券设计制版有限公司 | Image processing system based on multi-level image location and method |
CN104463171A (en) * | 2014-12-09 | 2015-03-25 | 四川大学 | Seal inscription extraction method based on PCNN |
CN104463122A (en) * | 2014-12-09 | 2015-03-25 | 四川大学 | Seal recognition method based on PCNN |
CN104637055B (en) * | 2015-01-30 | 2018-09-28 | 天津科技大学 | A kind of high precision image matching process based on small scale features point |
CN106709490B (en) * | 2015-07-31 | 2020-02-07 | 腾讯科技(深圳)有限公司 | Character recognition method and device |
CN105410988B (en) * | 2015-12-17 | 2017-05-31 | 中国电子科技集团公司第四十一研究所 | A kind of cigarette stealth coding integrity detection device |
CN105825211B (en) * | 2016-03-17 | 2019-05-31 | 世纪龙信息网络有限责任公司 | Business card identification method, apparatus and system |
JP2018088116A (en) * | 2016-11-29 | 2018-06-07 | キヤノン株式会社 | Information processing device, program and information processing method |
CN108255901B (en) * | 2017-04-06 | 2019-03-08 | 新华智云科技有限公司 | A kind of paper media's web page interlinkage sending method and equipment |
CN108804499B (en) * | 2018-04-03 | 2022-04-29 | 南昌奇眸科技有限公司 | Trademark image retrieval method |
CN108763261B (en) * | 2018-04-03 | 2021-08-27 | 南昌奇眸科技有限公司 | Graph retrieval method |
CN108845998B (en) * | 2018-04-03 | 2022-04-29 | 南昌奇眸科技有限公司 | Trademark image retrieval and matching method |
CN109493369B (en) * | 2018-09-11 | 2020-12-29 | 深圳控石智能系统有限公司 | Intelligent robot vision dynamic positioning and tracking method and system |
CN109543770A (en) * | 2018-11-30 | 2019-03-29 | 合肥泰禾光电科技股份有限公司 | Dot character recognition methods and device |
CN109744996B (en) * | 2019-01-11 | 2021-06-15 | 中南大学 | OCT image BMO position positioning method |
CN110097046A (en) * | 2019-03-11 | 2019-08-06 | 上海肇观电子科技有限公司 | A kind of character detecting method and device, equipment and computer readable storage medium |
CN110632094B (en) * | 2019-07-24 | 2022-04-19 | 北京中科慧眼科技有限公司 | Pattern quality detection method, device and system based on point-by-point comparison analysis |
CN114078109A (en) * | 2020-08-13 | 2022-02-22 | 鸿富锦精密电子(天津)有限公司 | Image processing method, electronic device, and storage medium |
CN114266908A (en) * | 2021-12-17 | 2022-04-01 | 福建顺景机械工业有限公司 | Character screening method based on image pixel value characteristics |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101184167A (en) * | 2007-12-14 | 2008-05-21 | 西安交通大学 | Industry intelligent camera |
CN101276218A (en) * | 2008-04-29 | 2008-10-01 | 上海交通大学 | Machine vision system aiming at high real time image collecting and processing |
CN101315664A (en) * | 2008-05-27 | 2008-12-03 | 昆明理工大学 | Text image preprocessing method for character recognition |
-
2009
- 2009-05-11 CN CN2009103022001A patent/CN101576956B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101184167A (en) * | 2007-12-14 | 2008-05-21 | 西安交通大学 | Industry intelligent camera |
CN101276218A (en) * | 2008-04-29 | 2008-10-01 | 上海交通大学 | Machine vision system aiming at high real time image collecting and processing |
CN101315664A (en) * | 2008-05-27 | 2008-12-03 | 昆明理工大学 | Text image preprocessing method for character recognition |
Also Published As
Publication number | Publication date |
---|---|
CN101576956A (en) | 2009-11-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101576956B (en) | On-line character detection method based on machine vision and system thereof | |
CN110288032B (en) | Vehicle driving track type detection method and device | |
CN110427793A (en) | A kind of code detection method and its system based on deep learning | |
CN106529559A (en) | Pointer-type circular multi-dashboard real-time reading identification method | |
CN116188475A (en) | Intelligent control method, system and medium for automatic optical detection of appearance defects | |
CN112149667A (en) | Method for automatically reading pointer type instrument based on deep learning | |
CN103530590A (en) | DPM (direct part mark) two-dimensional code recognition system | |
CN105303169B (en) | A kind of cell division identification method and its identification device based on slow feature | |
CN106874901B (en) | Driving license identification method and device | |
CN111126393A (en) | Vehicle appearance refitting judgment method and device, computer equipment and storage medium | |
CN103854278A (en) | Printed circuit board image registration method based on shape context of mass center of communicated region | |
CN111583180A (en) | Image tampering identification method and device, computer equipment and storage medium | |
CN113657339A (en) | Instrument pointer counting and reading method and medium based on machine vision | |
CN114662594B (en) | Target feature recognition analysis system | |
CN117523214A (en) | Method for intelligently identifying anti-fraud of picture modification trace | |
CN105139508A (en) | Method and device for detecting paper money | |
CN110728269A (en) | High-speed rail contact net support pole number plate identification method | |
CN116704270A (en) | Intelligent equipment positioning marking method based on image processing | |
CN113591863A (en) | License plate character recognition method with self-correction consciousness | |
CN114998704A (en) | Yolov 5-based flaw target detection method | |
CN113920450A (en) | Method and device for identifying insulator RTV coating based on intrinsic image decomposition | |
Kesiman et al. | Benchmarking A New Dataset for Coffee Bean Defects Classification Based on SNI 01-2907-2008 | |
CN118397316B (en) | Track train item point missing detection method | |
Tao et al. | A hybrid approach to detection and recognition of dashboard information in real-time | |
CN117282687B (en) | Automatic mark picking and supplementing control system for visual inspection of printed matter |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110831 Termination date: 20180511 |